Use Cases

SportsAgent · Sports information

Ask about the matches.
Get a grounded answer.

Players scrolling through fixture lists, switching tabs for team form, jumping out to check who's injured — that's friction. SportsAgent is a conversational interface over the platform's live data: ask about upcoming matches, tournaments, team form, head-to-head history, or lineup news in plain language and get an answer drawn from the platform's own catalogue and statistics — never from model memory, never from the open internet.

LLM MCP Streaming Multi-language

Overview

SportsAgent gives a sports information platform's users a conversational interface to the catalogue. One question, one streamed answer, drawn from the platform's own data — no scrolling between fixture lists, statistics pages, and team profiles.

Capabilities

What it does

18:45 Pera

Who's missing for Liverpool?
Two midfielders are out — one calf, one match-fitness.

Conversational interface over live fixtures, tournaments, teams, and statistics

One question, three lookups

Resolve tournament
Find this weekend's fixture
Pull form + head-to-head

Chains tool calls — resolves tournaments and teams, finds fixtures, pulls stats — to compose one grounded answer

One agent, four languages

EN DE + 2 more

Multi-language out of one configuration — the user's UI language is passed per prompt; the assistant answers in that language

Sports Assistant is thinking

Real-time streaming — the user sees the answer arriving token by token, the modern chat-product feel, no waiting for the full response

Integration

SportsAgent lives inside the platform via a single embed — no infrastructure change. Dokko's agent layer talks to a read-only MCP server the platform stands up over its own APIs, so every factual claim is grounded in the platform's live data, never composed from model memory. Per-user context (scope, timezone) flows through opaquely so the model never sees raw identifiers.

Editorial control

The platform stays in charge of what the assistant says.

The assistant runs inside a thin platform-side proxy that brackets every conversation with the platform's own guardrails. Inputs are checked before they reach Dokko — prompt-injection patterns, off-topic content, profanity, length caps. Outputs are checked per streamed chunk, with the platform able to allow, redact, truncate, or abort. The assistant's scope is locked in plain-language agent instructions versioned in Dokko — no general sports news, no editorial guidance, no predictions or recommendations. The integration surface is designed for this: discrete prompt submission and a typed event stream give the platform clean interception points on both sides, so domain policy stays entirely in the platform's hands.

Get started

Read deep. Act fast.
On your documents.

The 14-day trial runs on your real library, not a sandbox. Bring your contracts, manuals, or frameworks — and see how Dokko reads them.